Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75899.47831828266,
-      "required_pages_per_second_upper_estimate": 0.005250902021111861
+      "required_token_positions_per_second": 77636.33772862358,
+      "required_pages_per_second_upper_estimate": 0.005371061985188313
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9487.434789785333,
-      "required_pages_per_second_upper_estimate": 0.003972190826315713
+      "required_token_positions_per_second": 9704.542216077947,
+      "required_pages_per_second_upper_estimate": 0.004063089171985823
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18974.869579570666,
-      "required_pages_per_second_upper_estimate": 0.0039519649324632495
+      "required_token_positions_per_second": 19409.084432155894,
+      "required_pages_per_second_upper_estimate": 0.004042400435241042
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28462.304369356,
-      "required_pages_per_second_upper_estimate": 0.0017062716875205024
+      "required_token_positions_per_second": 29113.626648233843,
+      "required_pages_per_second_upper_estimate": 0.0017453174636276932
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28462.304369356,
-      "required_pages_per_second_upper_estimate": 0.0022128189722712536
+      "required_token_positions_per_second": 29113.626648233843,
+      "required_pages_per_second_upper_estimate": 0.0022634564145900694
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28462.304369356,
-      "required_pages_per_second_upper_estimate": 0.365083880008671
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29113.626648233843,
+      "required_pages_per_second_upper_estimate": 0.37343834286673905
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11211643608595523,
-    "producer_p95_seconds": 0.31587266200222075,
-    "gpu_update_seconds": 0.9713900758424633,
-    "producer_batches_per_second": 8.919298854927387,
-    "gpu_batches_per_second": 1.0294525596555266,
-    "producer_headroom": 8.66411839114951,
+    "producer_batch_seconds": 0.14946511808625473,
+    "producer_p95_seconds": 0.33449703899532324,
+    "gpu_update_seconds": 0.9496583965322385,
+    "producer_batches_per_second": 6.690524269501534,
+    "gpu_batches_per_second": 1.0530102230987357,
+    "producer_headroom": 6.353712549734853,
     "sustainable": true,
-    "buffer_seconds": 60.22618470223272,
+    "buffer_seconds": 60.77813737806326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 506052.1489869873,
+    "generation_bytes_per_second": 543379.7397463934,
     "upload_bytes_per_second": 4989679.253709886,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.58701088149274,
+    "estimated_full_shards_per_hour": 116.59664291662075,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -377,3 +377,4 @@
 {"step": 682700, "loss": 0.510818362236023, "lr": 0.0001, "grad_norm": 0.20056314603891742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99929748758, "tokens_per_second": 153672.10828742673, "data_wait_s": 1.0219999239780009e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 682800, "loss": 0.5299861431121826, "lr": 0.0001, "grad_norm": 0.19779118532524445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99944461786, "tokens_per_second": 153490.70288602065, "data_wait_s": 1.0031995770987123e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 682900, "loss": 0.5198776125907898, "lr": 0.0001, "grad_norm": 0.19194340221881184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99958976162, "tokens_per_second": 151873.5404881638, "data_wait_s": 1.4601995644625276e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 683000, "loss": 0.5247899889945984, "lr": 0.0001, "grad_norm": 0.20268397962436385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99973492914, "tokens_per_second": 151112.64794044953, "data_wait_s": 1.226499443873763e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:664b211013892024c19ed2d40fcc9146b5a9ca26c60f60c7709b2c282ad15b2c
-size 808368
+oid sha256:e4234fd2f238c404ded19b4cfc44c71cda9c83d137a19eaeb05911c0aa479c95
+size 810486